Upload
truongdieu
View
213
Download
0
Embed Size (px)
Citation preview
Running title: EXTERNAL AND INTERNAL UNCERTAINTY
There is a 60% probability, but I am 70% certain: Communicative
consequences of external and internal expressions of uncertainty
Erik Løhre1, 2 and Karl Halvor Teigen1, 2
1Simula Research Laboratory, Oslo
2 Department of Psychology, University of Oslo
Corresponding author:
Erik Løhre
Simula Research Laboratory
P.O. Box 134, 1325 Lysaker, Norway
Telephone: +47 41 42 01 35
E-mail: [email protected]
Word Count: 10338 words
1
Abstract (169 words)
Current theories of probability recognize a distinction between external (un)certainty
(frequentistic probabilities) and internal (un)certainty (degrees of belief). The present studies
investigated this distinction in lay people’s judgments of probability statements formulated to
suggest either an internal (“I am X% certain”) or an external (“It is X% certain” or “There is a
X% probability”) interpretation. These subtle differences in wording influenced participants’
perceptions and endorsements of such statements, and their impressions of the speaker.
External expressions were seen to signal more reliable task duration estimates, and a lower
degree of external than internal certainty was deemed necessary to advise a course of action.
In conversations about football, internal expressions were perceived as signaling more
personal interest, and were expected to be on the average 10% higher than corresponding
external probabilities. Finally, people who reported their outcome expectations for two major
sports events let their degree of interest in these events influence their internal but not their
external certainty. These results have implications for the communication of uncertainty and
probability.
Keywords:
Internal uncertainty; external uncertainty; communication; subjective probability; predictions
2
There is a 60% probability, but I am 70% certain: Communicative consequences of
external and internal expressions of uncertainty
People spend much of their time trying to predict future outcomes both in their private
lives and in professional settings (Dunning, 2007), but by doing so they enter a domain
fraught with considerable uncertainty. A football supporter who asserts that Barcelona will
win the Champions League next year, should admit that the probability for this scenario is
less than 100%. The uncertainty of a prediction can be communicated in different ways: with
numerical probabilities (there is a 70% chance, I am 70% certain), with verbal probability
expressions (it is rather likely, I am almost sure), or by suggesting a range of possible
outcomes (Barcelona, Chelsea, or Bayern will win the Champions League). Ranges are often
combined with numerical or verbal probability statements, for instance when climate experts
claim that “A global temperature increase of 2-4 degrees is likely (p > 66%)”. Moreover, a
narrow range of likely outcomes (2.5-3.5 degrees) suggests higher certainty than a wider
range.
The way in which a speaker communicates uncertainty can influence others’
decisions. For instance, when a doctor says that it is “quite uncertain” that a treatment will be
helpful, people are less willing to recommend the treatment than if the doctor says that there is
“some possibility”, even though both expressions are believed to indicate the same probability
(around 35%) of success (Teigen & Brun, 1999). In this case, people are swayed by having
the probabilistic message framed in negative or in positive verbal terms.
Compared to verbal probabilities, numerical probabilities are more precise, and may
be the result of well-defined calculations, as when a 70% probability for a target outcome T
implies that T will occur in 7 out of 10 cases. However, even numerical probabilities can give
rise to divergent interpretations. For instance, a 70% probability will be perceived as more
3
likely for expected outcomes than for unexpected outcomes (Windschitl & Weber, 1999), and
a weather forecast of “30% chance of rain tomorrow” can be misinterpreted to mean that “it
will rain 30% of the time” or “in 30% of the area” (Gigerenzer, Hertwig, van den Broek,
Fasolo, & Katsikopoulos, 2005).
In the present paper, we argue that people’s use and understanding of numerical
probabilities are also influenced by the assumed or expressed source of uncertainty. More
specifically, a probability statement can be seen as reflecting (1) internal uncertainty, based
on the speaker’s personal level of knowledge and degree of conviction, or (2) external
uncertainty, stemming from the operations of causal factors and random processes in the outer
world. We apply this distinction to predictions of different kinds, which are easily seen to
involve both internal and external uncertainty. Both the uncertainty term (confidence vs.
probability) and the voice of the prediction (personal vs. impersonal) can serve to make one or
the other source of uncertainty more or less salient. We assume that a first person voice, like
“I am 90% certain”, suggests a personal, internal source of certainty or uncertainty, whereas
third person statements, like “It is 90% certain”, suggest a degree of (un)certainty that can be
attributed to impersonal, external sources.
The theoretical distinction between external and internal uncertainty
Ever since probability theory was first developed in the 17th century there has been a
continuous philosophical discussion of how probability should be conceptualized. Originally,
probability theory was developed in the context of games of chance, such as tossing a coin,
where each outcome is equiprobable from a physical and logical point of view, and also
empirically demonstrable by observing the long-run frequency of outcomes. For instance,
flipping a coin 100 times will produce heads about 50% of the time, confirming that a 50%
chance of heads can be regarded as an external, objective feature of coin tossing. But this
probability value can also be taken to signify an individual’s degree of belief regarding a
4
particular outcome. Such beliefs describe a subjective state of mind, a degree of (incomplete)
knowledge, rather than a random process, and cannot easily be given a frequentistic
interpretation. If you are asked whether Franklin D. Roosevelt or Benjamin Franklin are
portrayed on the dime you are tossing, and you say without looking, “I am 80% sure that
Roosevelt is on the dime”, you indicate that you, personally, believe it is Roosevelt, but do not
feel completely certain about your guess, and not that 80% of the dimes to be tossed carry
Roosevelt’s face.
Different theorists have referred to these two variants of uncertainty as aleatory vs.
epistemic (Hacking, 1975), ontological (ontic) vs. epistemic (Taleb, 2010), objective vs.
epistemic (Gillies, 2000), and external vs. internal (Kahneman & Tversky, 1982). Kahneman
and Tversky also distinguish between different ways of assessing uncertainty. External
uncertainty, they argue, can be assessed either in the singular or in the distributional mode.
When the singular mode is used, probability is assessed by “the propensities of the particular
case at hand” – in other words, through an assessment of potential causal (facilitative or
inhibitory) factors involved. In the distributional mode, probability is assessed in a
frequentistic way, and the case at hand is seen “as an instance of a class of similar cases”.
Internal uncertainty, on the other hand, is assessed either through sifting and weighting of
arguments (“I do not believe that they would put Benjamin Franklin on the dime, since he is
on the 100 dollar bill”), or through an introspective judgment of the strength of a conviction
(“I seem to remember Roosevelt on the dime, but I am not completely sure”) (see also
Lagnado & Sloman, 2004).
In this paper, we will adopt the terminology of Kahneman and Tversky (1982) in their
distinction between internal and external uncertainty. However, both Kahneman and Tversky
and Fox and Ülkümen (2011) argue that the two variants of uncertainty are not mutually
exclusive, and that for many real-world events, probability estimates reflect a mixture of both.
5
For instance, if you are going to bet on a football match, your knowledge of the teams,
strengthened by your team loyalty, makes you feel 90% confident about Team A as the
winner. However, you also realize that every match contains unforeseeable events – injuries,
referee mistakes, weather conditions, and so forth. If this external uncertainty is taken into
account, you may end up giving Team A a 60% rather than a 90% chance.
External and internal uncertainty expressions in communication
The distinction between external and internal uncertainty is reflected in natural
language (Hacking, 1975; Kahneman & Tversky, 1982; Teigen, 1988). Fox and Ülkümen
(2011) summarized research showing how phrases suggesting different sources of uncertainty
are used by speakers and understood by listeners when engaged in conversations about
uncertain outcomes. Note that in Fox and Ülkümen´s view, epistemic uncertainty includes
both internal and external-singular uncertainty in Kahneman and Tversky´s terms, while
aleatory uncertainty includes only the external-distributional category. In this paper, we do
not directly address the distinction between singular and distributional assessments of
uncertainty, but our conception of external uncertainty agrees more closely with the original
proposal of Kahneman and Tversky, as we believe that uncertainty for singular events can
also be seen as “external” (for instance if probability is assessed through an evaluation of
causal propensities). Fox and Ülkümen identified expressions such as “I am 90% sure”, “I’m
reasonably confident” or “I am not sure” as signaling epistemic (internal) uncertainty,
whereas expressions like “I think there´s a 90% chance”, “I´d say there´s a high probability”
or “There is a small chance” signal aleatory (external) uncertainty. It follows that listeners
might draw different conclusions depending upon whether a probability is presented in one
way or the other.
Indeed, Fox and Malle (1997; cited in Fox & Irwin, 1998) found that when an expert
used an internal expression (“I am 60% sure that the team will win”), people were more
6
willing to bet on the suggested team than when an external expression (“I think there is a 60%
chance that the team will win”) was used. Fox and Irwin (1998) concluded that internal mode
statements were translated into higher “belief strength” than external mode statements.
Internal statements are person-dependent. Fox and Malle (1997) accordingly found
that internal statements led to more responsibility being attributed to the speaker. Participants
were more willing to promote an economist who made a correct prediction and more willing
to fire an economist who made an incorrect prediction phrased as an internal expression (I am
70% sure) compared to one who had expressed the same expectation in external terms (I think
there is a 70% chance).
Fox, Ülkümen, and Malle (2011, cited in Fox & Ülkümen, 2011) further found that
both speakers and listeners associate internal uncertainty with singular information, such as
the strength of an impression, while external uncertainty is associated with distributional
information, such as relative frequencies. For instance, predictions of the outcome of a
football match couched in internal language were associated with considerations of the
current strengths and weaknesses of the two teams, while external language was associated
with distributional reasoning about how often each team has won in the past.
However, the first part of the “external” expressions in the studies of Fox and
colleagues is arguably “internal”. “I think there is a 60% chance” seems to indicate that the
speaker believes that there is a 60% chance, and is not simply giving an impersonal report
based on objective frequencies (or other external sources of evidence). In other words, the
phrase may not be interpreted as purely external, but as a statement about one’s internal
uncertainty about an external uncertainty. This might have attenuated the objectivity of the
presumed external expressions. In the current studies internal expressions will be compared
with what we believe are more pure external expressions of uncertainty.
The present research
7
In their studies, Fox and colleagues contrast different uncertainty terms while keeping
the speaker’s voice (first person active) unchanged. We assume, in addition, that the same
term (e.g., “certain”) can point towards different sources of (un)certainty depending on the
voice. We assume that “impersonal” (third person) phrases, like “it is 90% certain” and “there
is a 70% probability”, point towards an external attribution of uncertainty.1 Such externally-
focused expressions could make people think of external sources of evidence (e.g. statistical
records), which lend these expressions an air of objectivity. This might in turn give such
statements an appearance of being valid and believable. On the other hand, first person
expressions, like “I am 90% certain” and “I am 70% confident”, point towards an internal
attribution of uncertainty. Such internally-focused statements are more speaker-dependent,
and can be supposed to vary between individuals. Hence they might be perceived as less valid
sources of information about objective chances, but more informative about characteristics of
the communicator, like the speakers’ beliefs and commitments. We use the terms “internally-
focused” and “externally-focused” to emphasize that the different expressions point toward
different sources of uncertainty, without excluding overlapping sources of evidence. For
instance, an internally-focused statement such as “I am 70% certain” might be based on
statistical information, but phrased in such a way as to reflect one’s personal beliefs.
In the following, we present five studies examining hypotheses derived from these
presumed characteristics. The first four experiments are based on fictitious predictions by
speakers using statements differing in voice (it is/there is vs. I am), and (occasionally) terms
(certain vs. probable). In Experiment 5 we ask respondents to state their own degree of
certainty of real events within an external or an internal frame. All comparisons were made
between subjects in different conditions (thus participants were not given the opportunity to
1 In statements of the form “It is 90% certain that Team A will win”, “it” is from a linguistic perspective regarded as an anticipatory subject, the logical subject being the embedded clause starting with “that...” (Kaltenböck, 2003). This could be rephrased to: “That Team A will win is 90% certain”, omitting the pronoun altogether. Such statements intend to describe external facts rather than the speaker’s (or someone else’s) beliefs and opinions.
8
make a direct comparison between these two modes, as they were asked to do in most of the
studies by Fox and colleagues).
Experiment 1 addresses the issue of perceived validity and believability. Fox and
Malle (1997) found that listeners preferred to bet on a team that the bookie said he was “60%
sure” would win, rather than on a team that another bookie said he thought had “a 60%
chance” of winning. We expected, in contrast, that speakers using third-person statements
would be seen as more knowledgeable, informative, and trustworthy, simply by using a more
objective language hinting at external rather than internal probabilities. Instead of considering
sports events, participants in this experiment were asked to evaluate predictions of task
durations. Such predictions are important both in daily life and in professional settings, and
are often surrounded by considerable uncertainty (Halkjelsvik & Jørgensen, 2012).
Experiment 1 also investigated how predictions containing internally- or externally-focused
expressions of uncertainty are evaluated after the outcome is known. As speakers in the
internally focused condition appear as more personally responsible for their predictions they
may be evaluated more strictly (blamed more strongly) when they are wrong. Inaccurate
externally-focused statements might be easier to excuse as being the result of random
influences that are hard to predict.
Assessments of uncertainty are important for advisors and decision makers. In
Experiment 2, participants were asked to state the degree of certainty needed to recommend a
particular course of action (e.g., a medical treatment), formulated to be either internally- or
externally-focused: “To recommend this treatment, I should be at least _____ % sure [there
should be at least a _____ % probability] of success”. If “external” expressions hint at
objective, valid sources of evidence, while “internal” expressions indicate that a personal
opinion is being expressed, we might expect a lower numerical probability to be sufficient for
recommending treatment in the external than in the internal mode.
9
Internally-focused expressions can be expected to carry more information about the
speaker than externally-focused expressions. In Experiments 3 and 4 participants were given
short conversations about football, and asked to which degree a speaker who says “I am X%
certain [There is a X% probability] that Team A will win” is a supporter of Team A and hopes
they will win. We hypothesize that internally-focused statements are perceived as more
supportive than externally-focused statements, and also that speakers can allow themselves to
express a higher level of certainty when using internally-focused expressions. Experiment 4
furthermore compared the two different “external” expressions used in the current studies (“It
is X% certain” and “There is a X% probability”) to see whether they differ from “internal”
expressions in the same way.
In the final two studies we leave the lab and ask people to assess retrospectively
(Experiment 5A) and prospectively (Experiment 5B) their own expectations of selected
outcomes of two actual sports events, namely the recent World Chess Championship and the
imminent Winter Olympic Games in Sochi. Participants in different conditions were asked to
answer these questions phrased with either an external (it was X% certain / it is certain) or an
internal focus (I was X% certain / I am certain). They were also asked to report their own
level of interest and engagement in these events. From the subjective character of internally-
focused expressions one might expect a higher degree of certainty from respondents who were
sufficiently engaged to have formed an opinion about outcomes. This was tested by checking
the correlations between measures of interest and degree of certainty for participants in the
two conditions. Personal involvement could be expected to predict degree of certainty
primarily in the internally-focused condition.
EXPERIMENT 1
10
Participants in the present experiment received interval predictions of task duration for
three hypothetical tasks, with two levels of internally- or externally-focused certainty. In
contrast to Fox and Malle (1997) who found that an expert using an internal expression was
preferred over an expert using an external expression, we hypothesize that externally-focused
expressions without the subjective phrase “I think” will be regarded as more informative, and
that the speaker will be seen as more knowledgeable and trustworthy when using an
externally- instead of an internally-focused expression. We further assume that higher degrees
of certainty are perceived as more informative than lower degrees, in line with the search for
definitive predictions principle proposed by Keren and Teigen (2001). These authors found
that probabilities closer to 100% were consistently perceived as more informative and more
useful than lower probabilities, but did not ask whether they also were more trustworthy.
In addition, the experiment addressed potential consequences of different expressions
for predictions that turn out to be inaccurate. Probabilistic intervals allow for occasional
misses. A 70% interval predicts in fact that out-of-range outcomes will occur in 3 out of 10
cases. This may be more obvious for expressions suggesting a frequentistic interpretation,
than for expressions reflecting degrees of belief, where an inaccurate prediction suggests that
that the belief was mistaken. Additionally, as internally-focused expressions allot more
responsibility to the speaker (Fox & Malle, 1997), one might expect that inaccurate
predictions will make listeners less willing to trust the speaker in the future.
Method
Participants. The experiment was posted on MTurk (Goodman, Cryder, & Cheema,
2013; Paolacci, Chandler, & Ipeirotis, 2010) and 289 American participants were recruited.
After excluding those who failed an instructional manipulation check (IMC; Oppenheimer,
Meyvis, & Davidenko, 2009), 225 participants (99 female, 126 male) were left for analysis.
The mean age was 32.4 years (one participant did not indicate age), and 76.0% of the
11
participants reported “Some college” or “College graduate (4 years)” as the highest completed
level of education. Participants were randomly assigned to one of four conditions.
Questionnaires. The participants were instructed to read three scenarios carefully and
to imagine that the scenarios happened to them. All scenarios described a person estimating
time. For instance, Scenario 1 described a hypothetical car trip:
You are going to drive your car from City A to City B. You ask your friend Mary
about the travel time. After thinking about it for a while, Mary says: “I am [It is] 90%
certain that it will take between 3 and 4 hours to drive from City A to City B.”
The other two scenarios described a person estimating the time required for a paint
job, and another person estimating the time it would take to make a dish from an online
recipe. The focus of the uncertainty expressions (internal: I am X% certain, external: It is X%
certain) and the level of certainty (high: 90%, low: 70%) in the statements were varied
between subjects in a 2 x 2 design.
Each scenario was followed by ratings of the speaker’s perceived knowledge and
trustworthiness, as well as the informativeness of the statement. For example, for Scenario 1,
participants were asked about their agreement with the following statements: (1) Mary seems
very knowledgeable about the travel time from City A to City B; (2) I trust Mary´s estimate;
and (3) Mary´s statement is very informative. All statements were to be rated on seven-point
Likert scales from 1 (Completely disagree) to 7 (Completely agree).
On the next page of the questionnaire participants were informed that each of the tasks
actually took somewhat more time than originally predicted. For Scenario 1, they were told:
You asked your friend Mary how long it would take to drive from City A to City B,
and she answered: “I am [It is] 90% certain that it will take between 3 and 4 hours to
drive from City A to City B.” Imagine that it actually took 4.5 hours to drive from
City A to City B.
12
Based on this outcome information, participants were asked to rate how surprised they
were about the actual duration, to what extent the original statement agreed with the facts, and
to what extent they would trust the speaker to estimate a similar task at a later occasion. For
Scenario 1, the statements were: (1) I am very surprised that it actually took 4.5 hours to drive
from City A to City B; (2) I think Mary´s statement agrees very well with the facts; and (3) I
would trust Mary the next time she estimates how long it takes to drive somewhere.
<Table 1 about here>
Results
Ratings were affected both by level of certainty and focus of expressions, as shown in
Table 1, and confirmed by separate 2 x 2 ANOVAs. High certainty predictions were
considered more informative, more to be trusted, and implied more knowledge, and
engendered also more surprise than lower certainty predictions when they failed (all p´s
< .005). More interestingly, there were also main effects of the focus of expressions
(summarized in Table 1). Speakers who used externally-focused (impersonal) expressions
were perceived as more knowledgeable, and their statements appeared more informative than
speakers who used internally-focused (personal) expressions. There was further a slight
tendency to trust externally-focused statements more. After getting outcome information,
participants perceived externally-focused statements also to agree somewhat better with the
facts, and showed a tendency to place more future trust in speakers who used “external”
expressions. Ratings of knowledge, trust, and informativeness were highly correlated (r´s
of .73 to .85) within the four conditions. Also, judgments of accuracy predicted future trust,
with high correlations (r’s of .80 to .90) within each of the four conditions.
Discussion
13
The pre-outcome ratings showed that speakers’ interval estimates are regarded as more
informative, and the speakers are perceived as more knowledgeable, when describing their
degree of certainty in impersonal rather than personal terms. This effect was obtained even
though the manipulation of the focus of uncertainty expression was very subtle (I am X%
certain vs. It is X% certain). There was no indication for internally-focused statements to be
trusted more than externally-focused statements, as Fox and Malle (1997) reported, but rather
the other way around. These results indicate that listeners think that statements in the third
person voice, which indicate external certainty, are more valid than first-person statements.
The effect was not found to be dependent upon of the level of certainty (the ANOVAs
revealed no significant interactions, all p’s > .185). After feedback is given and it is revealed
that the speaker´s predictions were too optimistic, there was a tendency to find externally-
focused statements in better agreement with the facts, and to trust new predictions by the
same speaker. This also follows from the hypothesis that greater personal responsibility is
attached to internally-focused probabilities.
EXPERIMENT 2
If internally-focused expressions suggest that the speaker is stating his or her personal
opinion, while externally-focused expressions suggest objective sources of evidence,
statements containing impersonal estimates of certainty should also form a better basis for
recommendations and decisions under uncertainty. In Experiment 2, participants were asked
to imagine an advisor who informs them of the level of certainty needed to recommend a
particular action. In this experiment the focus of uncertainty expressions was manipulated
both by voice (first vs. third person) and term (I am X% sure vs. it is Y% probable). The
hypothesis is that people are willing to accept a lower value of Y than of X. This follows from
a perception of “external” probabilities (Y) as more reliable, as suggested by Experiment 1. It
14
is also compatible with the higher degree of responsibility attached to statements of personal
certainty. It seems reasonable that an advisor who might be made responsible for his or her
recommendations would like to be pretty sure.
Method
Participants. The survey was posted on MTurk and 139 people responded. After
excluding 4 people that failed an attention check, 135 American participants (44 female, 91
male) were retained for analysis. The mean age was 30.3 years (SD = 8.7 years), and 74.8% of
the participants indicated “Some college” or “College graduate (4 years)” as the highest
achieved degree of education.
Questionnaires. The survey contained three scenarios, given to participants in one
internally- and another externally-focused condition. The first two scenarios described people
who asked for advice on whether or not to choose a course of action. For example, the first
scenario described Jessica, who is suffering from migraine (adapted from Teigen & Brun,
1999):
Jessica has periodically been suffering from migraine headaches, and is now
considering a new method of treatment based on acupuncture. The treatment is rather
costly and long-lasting. Jessica asks if you think she should give it a try. Fortunately
you happen to know a physician with good knowledge of migraine treatment, whom
you can ask for advice. After thinking about your question, the physician says: “To
recommend this treatment, I should be at least _____ % sure [there should be at least a
_____ % probability] that it will be helpful for Jessica.”
The second scenario described Michael, a student preparing for an exam who is asking
for advice on whether or not he should read a book that was not discussed in the lectures, in
order to have a possibility to impress the examiners.
15
The questionnaire included also a third scenario on a different issue, namely whether
new information affects internally- or externally-focused expressions in different degrees.
This scenario (adapted from Juanchich, Teigen, & Villejoubert, 2010) described a profiler
working on a criminal case, who concludes that based on the evidence, there is a 30%
probability [he is 30% sure] that a suspect is guilty. Then new evidence is found pointing in
the direction of the original suspect, which makes it more probable [makes the profiler more
sure] that the suspect is guilty. The participants were asked to complete a statement attributed
to the profiler indicating how probable it is, or how sure the profiler is, that the suspect is
guilty, given this new information.
<Table 2 about here>
Results
As shown in Table 2, the minimum probability for recommending a treatment for
Jessica and advising Michael to read an extra-curricular book was around 60% in the
externally-focused condition, whereas the minimum personal certainty (internally-focused
condition) had to be on the average about 10 points higher, around 70%. Thus, people seemed
to think that the external probability needed to recommend an action is lower than the
required internal, subjective level of confidence. The effect was detected across two scenarios
describing quite different events.
There was no difference in how internally- and externally-focused statements were
completed in the third (profiler) scenario. New evidence led to the same increase in certainty
as in probability estimates.
EXPERIMENT 3
16
The first two experiments indicate that externally-focused expressions are considered
more predictive than internally-focused expressions about events in the external world. But
internally-focused expressions may be more informative about the speaker’s attitudes and
preferences. If you say that you are 90% certain that a team will win, you might be hoping
more strongly that the team will win than if you say that there is a 90% probability. In
addition, an internal focus might also indicate that you are basing your statement on your
intuition rather than on facts or statistics. These two hypotheses were investigated in
Experiment 3, where participants were asked to take the role of listeners to brief dialogues
about football matches. The focus of uncertainty expressions was again manipulated by
changing both term and voice, with “I am X% certain” representing internal and “there is a X
% probability” representing external focus.
Method
Participants. Students at the Department of Informatics at the University of Oslo
were approached during a break in a lecture. After excluding ten incomplete questionnaires,
there were 100 participants (25 female, 75 male) left for analysis. The mean age was 23.2
years (SD = 4.76, one participant did not indicate his age). Participants were randomly
allocated to four different conditions.
Questionnaires. The questionnaires described four brief conversations about football
matches that were actually going to be played about one month later. To give one example
(translated from Norwegian):
Fredrik and Markus discuss the matches in the first round of the Norwegian Premier
League. Fredrik says: “I guess Rosenborg wins over Viking”. Markus says: “There is a
90% probability that Rosenborg will win this match”.
Participants were asked to rate the second speaker according to (1) how much Markus
hopes that Rosenborg will win; (2) how much knowledge Markus has about football; and (3)
17
to which degree Markus’ statement appears to be based on his gut feelings (all ratings on 1-7
Likert scales).
The questionnaires differed according to a 2 x 2 design with focus of expression:
internal (“I am 90% certain…”) vs. external (“There is a 90% probability…”)2, and level of
certainty: high (80% or 90%) vs. low (60% or 70%), as the two factors.
Finally, all participants rated their interest in football and in the premier league on two
1-7 Likert scales. The scales were highly correlated, r = .70, and accordingly combined into
an overall interest scale.
<Table 3 about here>
Results
Mean ratings are shown in Table 3. Speakers using internally-focused expressions
were perceived as more supportive of the team (M = 4.58, SD = 1.46) than speakers using
externally-focused expressions (M = 3.91, SD = 1.53). A 2 x 2 ANOVA with focus of
uncertainty expression (internal vs. external) and level of certainty (high vs. low) as between-
subjects factors reveal a main effect of focus, F(1,96) = 4.936, p = .029, η2p=.049, but no
effect of level of certainty. There were no significant effects of any of the factors on ratings of
knowledge and gut feelings, and no interactions, all F´s < 1.
Overall, the participants indicated a rather low (M = 2.29, SD = 1.63) interest in
football. For unknown reasons, and despite being randomly assigned to the four conditions,
participants in the low internally- and high externally-focused conditions happened to be more
interested in football than participants in the other two conditions. However, including interest
as a covariate in the analysis did not change the main pattern of results, indicating that
differences in interest did not substantially affect other ratings.
2 In Norwegian, ”Jeg er 90% sikker” and ”Det er en 90% sannsynlighet”.
18
In contrast with Experiment 1, externally-focused expressions did not suggest more
knowledgeable speakers, nor were internally-focused expressions more strongly linked to gut
feelings, as we had expected. Perhaps the use of precise numerical values in all conditions
overshadowed the more subtle distinction between internal and external focus. It is also
possible that participants did not find the use of impersonal probabilities (for results of single
games in the beginning of the football season) particularly convincing. With these potential
explanations of the null-findings regarding ratings of knowledge and gut feelings in mind, it is
remarkable that the first person voice still clearly suggested a more hopeful speaker
EXPERIMENT 4
In the previous experiments, (un)certainty expressions with an internal focus (“I am X
% certain”) have been compared with two different expressions with an external focus,
namely “It is X% certain” and “There is a X% probability”. While the first differs from
internally-focused expressions only in voice (3rd vs. 1st person), the second also differs with
respect to uncertainty term (probability vs. certainty). Experiment 4 was set up to test whether
these two externally-focused (un)certainty expressions are interchangeable, and whether both
differ from internally-focused expressions in the same way. Ideally, this suggests a 2 x 2
design with voice and term as two independent factors. However, whereas the term “certain”
can be meaningfully combined with first person as well as third person personal pronouns,
this is not the case with “probable”. To say that “I am 70% probable” does not make sense.
Personal probabilities have in this case to be rewritten in indirect ways, mixing up internal
and external focus, as in phrases like “I believe there is a 70% probability”, or “my
probability is 70%”. To avoid such ambiguous constructions, we settled for a comparison
between three conditions: one internally-focused (I am X% certain), and two externally-
focused based on different terms (It is X% certain; There is a X% probability).
19
We expected the internally-focused statements to be associated with higher numerical
probabilities than the other two conditions (similar to Experiment 2), and also to indicate
greater personal involvement (similar to Experiment 3). The contradictory results for ratings
of knowledge in Experiment 1 and Experiment 3 did not allow any clear predictions for the
relationship between perceived knowledge and focus of uncertainty expressions. However, we
expected externally-focused expressions to be seen as at least equally if not more indicative of
knowledge as compared to internally-focused expressions.
Method
Participants. Psychology students attending two lectures at the University of Oslo
were invited to complete the questionnaires during lecture breaks. Of the 150 collected
responses, 10 incomplete questionnaires were discarded. This left 140 participants (95 female,
44 male, 1 did not indicate gender) with a mean age of 22.5 years (SD = 5.65, two participants
did not indicate age). The participants were randomly assigned to one of three different
conditions.
Questionnaires. The participants in this experiment were again asked to take the role
as listeners to conversations about football, but in this case, the conversations were about the
relegation battle in the Norwegian premier league toward the end of the 2014 season. At this
point, it made sense to discuss the chances of four teams with standings towards the bottom of
the Norwegian premier league table to stay in the premier league for the next season vs.
becoming relegated to a lower division. Another difference from Experiment 3 was that
participants were asked to fill in the level of certainty they thought the speaker would express.
For example:
Olav and Elin are discussing football while having lunch. Olav says: “Bodø/Glimt did
not have a good season this year”. Elin says: “I am _____ % certain that Bodø/Glimt
will keep their place in the premier league”.
20
There were three different versions of the questionnaires, with the focus of uncertainty
expressions varied between subjects: “I am _____ % certain…” (internal condition, n = 52),
“It is _____ % certain…” (external certainty condition, n = 46), and “There is a _____ %
probability…” (external probability condition, n = 42)3. The participants were instructed to
complete the statements in each conversation with appropriate numbers between 0 and 100.
Next, as in Experiment 3, they rated (1) to which degree the second speaker seems to support
the team in question, (2) how knowledgeable the speaker appears to be about football, and (3)
to which degree the statement seems to be based on gut feelings, and in addition rated their
interest in football and in the Norwegian premier league. Responses to the two interest
questions were highly correlated, r = .73, and were combined into an overall interest scale.
All ratings in the experiment were given on scales from 1 to 7, with higher numbers
indicating more perceived support, higher knowledge, more reliance on gut feelings, and
higher interest.
At the time of data collection, with 6 matches in the 2014 season remaining, the teams
discussed in the conversations were ranked as no. 12 to no. 15 out of 16 teams4. To ensure
that all participants had sufficient information to complete the questionnaires, the bottom part
of the premier league table after 24 of 30 matches was presented directly above the four
conversations, and the points system was briefly explained on the first page of the
questionnaire.
<Figure 1 about here>
Results
3 In Norwegian. “Jeg er _____% sikker”, “Det er _____% sikkert” and “Det er _____% sannsynlig”.4 A statistical model developed by researchers at the Norwegian Computing Center (Norsk Regnesentral) indicated that the probability of the four teams of staying in the premier league at the time of data collection was 93.1% for Bodø/Glimt, 73.2% for Haugesund, 62.1% for Brann, and 28.6% for Sogndal.
21
Probabilities: Mean probability ratings for each of the four teams, ordered according
to rank, are displayed in Figure 1. A linear trend indicates that in all conditions, speakers were
expected to express a lower probability of staying in the league the lower the standing of the
team. Compared to the probabilities obtained from a statistical model (see Footnote 3),
participants underestimated the probabilities for the three highest placed teams, while they
overestimated the probability for the lowest placed team (Sogndal). The figure also shows that
the internally-focused condition differs from the two externally-focused conditions, which are
similar to each other. This impression was confirmed by analyzing the average of the
probability ratings with a one-way ANOVA, which showed an effect of condition, F(2,137) =
6.212, p = .003, η2p = .083. Planned contrasts revealed that participants in the internally-
focused condition completed the statements with higher ratings (M = 57.3, SD = 20.9) than
participants in the two externally-focused conditions, t(137) = 3.503, p = .001, while the
external certainty condition (M = 45.4, SD = 17.9) and the external probability condition (M =
46.5, SD =15.7) did not differ from each other, t(137) = -.290, p = .773.
Support: An ANOVA on the perceived support of the speakers showed an effect of
condition, F(2,137) = 6.620, p = .002, η2p = .088. Speakers using an internally-focused
expression were thought to be more supportive (M = 4.83, SD = 1.30) than speakers using an
external certainty (M = 4.21, SD = 1.02) or an external probability expression (M = 4.02, SD =
1.07), t(137) = 3.580, p < .001, while there was no difference between the two externally-
focused conditions, t(137) = .748, p = .455, as shown by planned contrasts.
To check whether this difference in perceived support was simply produced by the
higher probability ratings in the internally-focused condition, we performed a mediation
analysis using condition as the independent variable (with two levels: internally-focused vs.
the two externally-focused conditions), probability ratings as the mediator, and perceived
support as the dependent variable. Although a bootstrap analysis with 20,000 samples
22
(Preacher & Hayes, 2004) showed a clear indirect effect of condition on support through
probability ratings, 95% CI [.097, .535], the direct effect of condition on support remained
significant, p = .029. In other words, internally-focused (vs. externally-focused) expressions
yielded an effect on perceived support that could not be fully accounted for by higher
probability ratings. Estimated probability also correlated positively with support within each
condition, especially in the internal condition (r = .48; p < .001), but also in the external
certainty (r = .33, p = .024) and probability (r = .39, p = .011) conditions.
Knowledge: The speakers were perceived to have a similar degree of knowledge in the
internally-focused condition (M = 4.20, SD = 1.07) as in the external certainty condition (M =
4.33, SD = .78) and the external probability condition (M = 4.20, SD = .73), F < 1, despite the
“internal” speakers’ higher level of expressed certainty. Degree of certainty was actually
negatively correlated with knowledge in the internal condition, r = -.28, p < .05, but not in the
two external conditions. Thus we find again no support for the claim that internally-focused
statements are regarded as more trustworthy than externally-focused statements (Fox & Irwin,
1998).
Gut feelings: Ratings of gut feelings did not reveal any significant differences between
the internal condition (M = 4.39, SD = 1.42), the external certainty condition (M = 4.24, SD =
1.41), and the external probability condition (M = 4.34, SD = 1.24), F < 1.
Own interest in football: The average interest in football was quite low both in the
internal condition (M = 2.56, SD = 1.76), the external certainty condition (M = 2.28, SD =
1.81), and the external probability condition (M = 2.60, SD = 1.54), with no difference
between conditions, F < 1.
EXPERIMENT 5A
23
The previous experiments were all vignette studies, where participants were asked to
draw inferences about the expertise, involvement, and estimates of hypothetical speakers. In
the final two studies, we ask participants to produce their own predictions of two actual sports
events, featuring internally- or externally-focused (un)certainty expressions. In these
situations, people might allow their personal certainty estimates to exceed their estimates of
the objective (impersonal) chances, provided they have some degree of personal involvement
in the predicted events. Participants with no opinions of their own would not be expected to
boost their “internal” estimates. Thus we can predict a correlation between personal
knowledge/interest and certainty for internally-, but not necessarily for externally-focused
certainty estimates.
The results reported in Experiment 5A were collected as part of a larger study
conducted before, during, and after the World Chess Championship 2013 with the purpose of
investigating people’s causal attributions of different outcome scenarios (Nordbye & Teigen,
2014). This competition was played as a match over 10 rounds between the reigning
champion Vishy Anand and the young Norwegian challenger, Magnus Carlsen. It ended up
with a solid victory for Carlsen (6.5-3.5), who also had been the bookmakers’ favorite. The
event was closely followed by Norwegian media and attracted considerable attention in this
country, also from people with no prior interest in chess. We asked participants to recall, in
hindsight, what they thought about Carlsen’s chances of winning, either as an “external”,
impersonal probability, or as their “internal”, personal certainty.
Method
Participants were recruited by e-mail distributed to students at the Department of
Psychology at the University of Oslo, resulting in 101 completed responses on a web-based
questionnaire, 75.2% women, mean age 25.7 years. They were randomly assigned to an
internally- (n = 50) or an externally-focused condition (n = 51). The study was conducted
24
approximately three weeks after the championship, and participants could win a gift card of
500 NOK (€ 63).
Before completing an extensive questionnaire about various factors responsible for
various hypothetical outcomes of the match, participants were asked to rate their own interest
in chess on five-point scales ranging from 0: no interest to 4: very interested in chess.
Next, they were asked to try to remember the thoughts they had about the
championship prior to the match. Who did they think would win the match, Carlsen or
Anand? They were then asked to assess, in retrospect, Carlsen’s winning chances on a scale
from 0 to 100%. This question was phrased slightly different in the two conditions, namely:
(1) How certain were you at that time that Carlsen would win (internal focus); or (2) How
certain was it at that time that Carlsen would win (external focus). In other words, the
conditions differed only in the voice of the statements.
Results
Mean interest in chess was 1.36 (SD = 1.14) in the internally-focused and 1.45 (SD =
1.17) in the externally-focused condition. Only 3 participants in each group reported that they
were very interested in chess, whereas 12 participants in each group (24%) reported no
interest at all.
A majority of 84.2% claimed that they had believed that Carlsen would win.
Participants in both groups reported the same overall level of certainty, with means of 57.4%
and 57.1% in the internally- and externally-focused conditions, respectively. However,
participants with an above-average prior interest expressed higher “internal” than “external”
certainty, as predicted, whereas those who claimed none or a low prior interest scored lower
in “internal” than “external” certainty. A 2 x 2 ANOVA with Condition (internally-focused
vs. externally-focused) and Interest (high vs. low) as the two factors showed a significant
25
interaction, F(1, 97) = 5.991, p = .016, η2 = .0585. Thus, participants with a high prior interest
were more personally certain that Carlsen would win than those with low prior interest, but
they did not claim a higher “objective” certainty. Personal certainty scores were significantly
associated with interest, r = .411, p = .003, but for impersonal certainty scores no such
relationship could be detected, r = .090, ns.
The correlation between interest and internally- but not externally-focused certainty
suggests that personal involvement and personal beliefs influence the way people think about
their own degree of certainty. However, participants in this study were asked to assess their
certainty in retrospect. Studies of the hindsight bias (Fischhoff, 1975; Roese & Vohs, 2012)
show that events often seem more predictable in retrospect than before the outcomes are
known. The results needed accordingly to be replicated with a prospective study.
EXPERIMENT 5B
The difference between answers to the question how certain it was and how certain
you were found in Experiment 5A was based on retrospective assessments of a single event in
an area of low prior interest to our student respondents. The present experiment involved
prospective predictions of the 2014 Winter Olympic Games, an event that traditionally can
claim a widespread public interest in Norway.
We expected that participants would allow themselves to show a higher degree of
internally- than externally-focused certainty (in line with the results of Experiment 2 and 4),
and thus would agree more with first-person than third-person statements. Observe that in this
experiment, percentages showing degree of certainty were not used, participants were simply
requested to indicate whether they were certain or not. We further expected to observe a
5 Similar results were obtained when interest scores were split in a “zero interest” vs. a “some interest” group (ratings above 0), rather than a with a high/low median split; F(1, 97) = 6.214, p = .014, η2 = .060.
26
positive correlation between interest in winter sports and internally- but not externally-
focused degree of certainty, in line with the results from Experiment 5A.
Method
Participants. Invitations to take part in the survey were distributed at three different
locations. A link to the web-based questionnaire was given to students at the Norwegian
school of sports sciences during a lecture, to students at the faculty of social sciences at the
University of Oslo on campus or via email, and was posted on Facebook. Altogether 101
participants (56.9 % women, mean age 27.2 years) completed the relevant questions in the last
three days preceding the opening of the Winter Olympics in Sochi (Russia).
Questionnaires. The survey contained 20 statements about outcomes that could occur
during the Olympics. All of the statements described outcomes that were seen as not unlikely
to happen, and were based on predictions from Norwegian newspapers and from
infostradasports.com (an internet site giving predictions based on the performance of
participating athletes in the months and years before the Olympics). Ten statements focused
on outcomes involving Norwegian athletes, ten other statements concerned outcomes
involving non-Norwegian athletes. For instance, one statement focused on one of the favorites
in cross-country skiing, Petter Northug: “I am [It is] certain that Petter Northug will get at
least one individual gold medal in cross-country skiing”. The participants were asked to
indicate how much they agreed with each statement on a scale from 1 (Completely disagree)
to 5 (Completely agree). For half the participants, the statements included an internally-
focused expression (“I am certain that…”), while for the other half, the statements contained
an externally-focused expression (“It is certain that…”). Again, external/internal focus was
manipulated only through changing the voice of the statements. The hypothesis was that
participants would agree more strongly with the “internal” than with the “external” certainty
statements.
27
At the end of the questionnaire, participants were asked to rate their level of interest in
the Olympics and how closely they were going to follow the events on two seven-point scales,
with 1 indicating very low interest, and 7 indicating very high interest.
Results
Each participant expressed to what degree he or she agreed with 20 statements about
what would “certainly” happen during the Olympics. The participants in the internally-
focused condition expressed, as predicted, an overall stronger agreement (M = 3.29, SD = .56)
than participants in the externally-focused condition (M = 3.06, SD = .70). However, this
difference failed to reach the 5% level of significance by a two-tailed test, F(1,99) = 3.486, p
= .065, η2p = .034.
The two interest scales were highly correlated (r = .910, p < .001), and a combined
interest score showed a slightly above medium interest for the Olympics both in the “internal”
(M = 4.11, SD = 2.01) and the “external” (M = 4.36, SD = 1.88) group. Observe that the
slight, but insignificant difference between groups, F(1,85) = .350, p = .56, runs counter to the
certainty results, indicating that the difference in certainty between groups cannot be
explained by a corresponding difference in interest. The correlations between interest and
certainty in the two conditions are presented in Table 4. The table shows strong positive
correlations between interest and internally-focused certainty, as predicted, but not between
interest and endorsement of externally-focused certainty statements. This correlation was not
limited to statements about Norwegian athletes, but applied even more strongly to statements
about non-Norwegian athletes (r = .46, p < .001). In other words, the more interested a
participant was in the Olympics, the more he or she endorsed statements starting with “I am
certain…”, whereas agreement with “It is certain…”-statements did not depend on the
participants’ level of interest.
28
<Table 4 about here>
GENERAL DISCUSSION
Although the discussion about internal and external conceptions of probability is
almost as old as probability theory itself (Hacking, 1975), this is one of the first attempts to
study this distinction systematically in communication about uncertain outcomes. As such, a
quite broad approach has been used, with different kinds of expressions, different kinds of
outcomes, and different communicative consequences pursued in the different experiments.
External and internal focus in communication of (un)certainty was in the present
studies manipulated partly by use of personal pronoun, or voice (it is certain vs. I am certain),
and partly by manipulating voice and terms simultaneously (there is a X% probability vs. I
am X% sure). This is a contrast to previous studies by Fox and colleagues where the implied
source of uncertainty was manipulated by changing the term while keeping the voice constant
(e.g., “I am 60% sure” vs. “I think there is a 60% probability”). Although we support the idea
that some terms are more likely to be associated with external uncertainty and distributional
reasoning, while other terms are more associated with internal uncertainty and singular
reasoning, our studies show that the voice is perhaps more important. The same (internally-
focused, according to Fox and colleagues) term, certain, is interpreted in a fashion congruent
with an external or internal attribution of uncertainty depending on the voice in which it is
given.
The results reported in this paper support the notion of “external”, impersonal
expressions as more informative and more to be trusted than “internal”, personal expressions
(Experiment 1), probably because of their appearance of coming from an objective source.
This also makes a 60% probability a better basis for decision than a personal confidence of
the same magnitude (Experiment 2). However, this presupposes that an objective source can
29
be found. The externally-focused probability statements of football matches in Experiment 3
were not perceived as indicative of expertise, possibly because participants were not
persuaded that they had an objective source. However, in conversations about the relegation
battle towards the end of the football season, “external” expressions were seen as indicating a
similar degree of knowledge as “internal” expressions of a higher numerical magnitude
(Experiment 4). Experiment 5B suggested that it might also be easier to disagree with third
person certainty, when we feel such certainty is not warranted.
The present findings indicate that expressions of “internal” personal certainty might be
easier to agree with (Experiment 5B), and seem to be more strongly related to personal
interest and engagement (Experiments 5A and 5B)6. Speakers using an internally-focused
expression of certainty reveal their preferences (Experiments 3 and 4), but at the same time
we expect them to be pretty sure before they recommend a line of action to others
(Experiment 2). This may be related to the lower trustworthiness (Experiment 1), as well as to
the greater personal responsibility implied by such expressions. Thus our studies highlight
several different facets of the personal and subjective nature of “internal” certainty or
uncertainty. This focus is hardly surprising, as the internally-focused expressions always
include the personal pronoun I. However, past research on lay people’s use of verbal and
numerical probabilities has largely overlooked the internal/external distinction, which the
present studies demonstrate as important in the communication of uncertainty.
Our findings are in general agreement with the frameworks suggested by Kahneman
and Tversky (1982) and by Fox and Ülkümen (2011). In these frameworks, internal
uncertainty refers to single cases, numerical estimates of certainty are perceived as judgments
of confidence, and uncertainty is attributed to lack of knowledge. External uncertainty on the
other hand suggests a class of outcomes, numerical probability estimates are believed to
6 However, interest may not only be related to increased internal certainty. In an (unpublished) study of people´s predictions of Oscar winners, we found positive correlations between interest and certainty in both the internal and the external conditions.
30
reflect relative frequencies or event propensities, and uncertainty is attributed to stochastic
behavior. The current studies show that speakers and listeners using internally - and
externally-focused expressions of uncertainty in communication of uncertain outcomes are
sensitive to at least some of these differences. In addition, our studies suggest differences in
involvement that were not considered earlier.
The current findings partly contradict previous conclusions drawn by Fox and
colleagues. Specifically, results from Experiment 1 indicate that speakers who use externally-
focused expressions are perceived as more knowledgeable, whereas Fox and Malle (1997)
found that listeners had a preference for speakers using internally-focused expressions. This
contradiction might be due to a difference in wording. Fox and colleagues let their speakers
say “I think there is a X% probability” (italics added), while in the current studies speakers
say “It is X% certain” or “There is a X% probability”. This makes the externally-focused
expressions more objective, and in our opinion, more clearly external. We acknowledge that
there are good reasons to include the phrase “I think”. It is quite seldom in real life that one
knows the objective probability of an outcome. However, it does not seem realistic that
laypeople (or even experts), when communicating uncertainty, will always hedge their
estimates with “I think”, and thereby call attention to the subjective element in an otherwise
objectively phrased statement.
Participants in the present studies did not always associate externally-focused
expressions with expertise. In Experiment 3, we found no difference in perceived
knowledgeability between speakers using expressions with an external and internal focus. As
speakers in this study were predicting the outcome of the first football matches in the season,
we can speculate that the listeners were perhaps not convinced that there was enough data to
give specific “objective” probabilities ahead of the games. In Experiment 4, objective
probabilities should make more sense, since the speakers were predicting the chances of
31
different teams of staying in the league towards the end of the season. Although this
experiment also failed to demonstrate any difference in knowledgeability, it should be noted
that externally-focused expressions were seen as equal to internally-focused expressions of a
higher magnitude. Predictions of task duration ranges (as in Experiment 1) can presumably be
made with more precision and be based on prior experience or other sources of external
evidence. Thus we might conclude that internal mode statements are not always translated
into higher “belief strength” than external mode statements (as found by Fox and Irwin,
1998), nor are estimates of “external” certainty always to be trusted more (as in our
Experiment 1), as it will all depend on the predictability of the target issue, or in other words,
on the amount of external uncertainty involved. A simple test of this explanation would be to
compare externally- and internally-focused expressions of uncertainty in a “kind” versus in a
“wicked” environment (Hogarth, 2010). In a “kind” domain, where regularities can be
observed and recorded, external expressions might signal greater knowledge, but in a less
transparent domain, where objective information about chances is difficult to attain, this will
not necessarily be the case.
To summarize, the present studies indicate that statements containing internally-
focused expressions of uncertainty will be perceived as different from those containing
externally-focused expressions, for at least two reasons: (1) being person-dependent, they
may be viewed as more variable and hence less to be trusted; but also (2) as involving the
speakers more strongly, revealing their commitments and making them more accountable for
their claims. Third person statements of “external” uncertainty appear, in contrast, to be more
objective and accordingly more reliable, but also less informative of the speaker’s views. The
results from the present experiments suggest both processes to be at work, in some cases (as
in Experiment 2) even as alternative, or complementary, explanations of the same pattern of
results.
32
Implications for communication of uncertainty. Numerical probability estimates
have usually been regarded as more objective and less vague than verbal probabilities. The
current studies show that even numerical probabilities can give rise to different inferences and
interpretations depending on whether they are communicated in such a way that the
uncertainty is attributed to an internal or an external source. This means that if you would like
to communicate that you have as precise knowledge as possible about an uncertain issue,
using a third-person phrase implying external certainty might be a good choice. However, in
areas where objective probabilities are not believed to exist, using an external expression
might seem unwarranted and perhaps downright dishonest. The stock analyst that claims “It is
90% certain that this stock will go up tomorrow” might be questioned as to how he can know
this; by saying “I am 90% certain”, he would have acknowledged this to be his personal
evaluation.
In areas where external evidence is believed to exist, third-person phrases may appear
more convincing. In scientific discourse, impersonal descriptions of certainty/uncertainty are
the rule, even when not computed but based on conclusions drawn by individual scientists.
For instance, a report from the Intergovernmental Panel on Climate Change (IPCC) showed
no occurrences of the first person plural pronoun “we” (Fløttum, 2010). Rather than using
internally-focused expressions such as “we are highly confident”, externally-focused
expressions such as “there is high confidence” were used. The current studies suggest that this
might make listeners perceive the report as more trustworthy. However, using an internally-
focused expression would make the speaker seem more personally engaged. One study of
climate change communication did actually find that the use of we is more engaging and more
likely to increase intentions to act environmentally friendly (Haddad, Morton, & Rabinovich,
2011; cited in Fløttum & Dahl, 2011).
As already hinted upon, different expressions might be used strategically for
33
persuasion purposes. A politician trying to gather support for an uncertain prediction (“There
is a 90% probability that the bridge will cost under 2 million”) might choose to use an
externally-focused expression to make the prediction seem more objective, and also to avoid
personal responsibility if the prediction fails. On the other hand, if you want to receive credit
for a successful prediction, expressions with an internal focus will make you seem more
involved. Hence, the politician who says “I am 90% certain that the project will be finished by
the end of the year,” will probably get more recognition if it turns out to be right.
Future directions. This paper has demonstrated several important differences
between “internal” and “external” expressions of uncertainty in communication. However, we
believe that this is just one piece of a more complicated picture. For instance, as pointed out
by Teigen (1988), external probabilities can be attributed to chance factors (“true”
randomness, luck, chances) or alternatively to dispositions or potentialities inherent in nature
(propensities). This could mean that different kinds of externally-focused expressions might
have somewhat different connotations. “There is a 60% chance” might lead to a stronger
focus on chance or randomness, while “There is a 60% probability” might be associated with
an event´s propensity to occur. This might have consequences for how external uncertainty is
expressed by speakers and interpreted by listeners.
If internal uncertainty signals lack of knowledge while external uncertainty signals
stochastic processes, one might also expect that different expressions can lead to different
levels of psychological discomfort. It has been argued that internal uncertainty is less
disturbing than external uncertainty, because internal uncertainty due to lack of knowledge
can be reduced by seeking more information, while external uncertainty due to stochastic
processes cannot be reduced in the same way (Neace, Deer, Michaud, & Bolling, 2011).
External uncertainty could implicate that there is less personal control over the outcome, and
the best way to handle this kind of uncertainty is to try to mitigate any potential negative
34
consequences. Investigating whether internal and external focus in uncertainty expressions
produces different affective reactions could be an interesting topic for future studies.
Another important question not addressed in the current studies, is when and why
speakers choose to use “internal” or “external” expressions of uncertainty. Is there a general
preference for one kind of expression? Or are there some situations where internal focus is
preferred, while other situations call for externally-focused expressions? Do people use
internally-focused expressions to communicate personal involvement or support? Or do
people use externally-focused expressions when they want to avoid responsibility? Answers
to questions such as these would improve our understanding of communication of uncertainty.
Conclusion
Taleb, in his book “The Black Swan” (2010) argues that the distinction between what
he calls epistemic (internal) and ontological (external) uncertainty is a “distinction without a
difference”. His point is that although the distinction makes sense philosophically, in practice
all uncertainty is epistemic, as there are very few cases where one can know with any
precision the probability of a specific future event. Although this undoubtedly is important to
keep in mind, the present studies show that the distinction between internal and external
uncertainty makes a difference to speakers and listeners psychologically. If you are going to
communicate uncertainty to someone else, you should know that your choice of an expression
with internal or external focus can lead to differences in interpretation. And as a listener, you
may already know, at least implicitly, that a speaker who says she is 90% certain is not saying
the same thing as the speaker who says there is a 90% probability.
35
References
Dunning, D. (2007). Prediction: The inside view. In A. W. Kruglanski & E. T. Higgins (Eds.),
Social psychology: Handbook of basic principles (pp. 69-90). New York: Guilford
Press.
Fischhoff, B. (1975). Hindsight ≠ foresight: The effect of outcome knowledge on judgment
under uncertainty. Journal of Experimental Psychology: Human Perception and
Performance, 1, 288-299. doi: 10.1037/0096-1523.1.3.288
Fløttum, K. (2010). A linguistic and discursive view on climate change discourse. La revue du
GERAS. ASp, 58, 19-37.
Fløttum, K., & Dahl, T. (2011). Climate change discourse: Scientific claims in a policy
setting. Fachsprache, (3-4), 205-219.
Fox, C. R., & Irwin, J. R. (1998). The role of context in the communication of uncertain
beliefs. Basic and Applied Social Psychology, 20, 57-70. doi:
10.1207/s15324834basp2001_6
Fox, C. R., & Malle, B. F. (1997). On the communication of uncertainty: Two modes of
linguistic expression. Unpublished manuscript.
Fox, C. R., & Ülkümen, G. (2011). Distinguishing two dimensions of uncertainty. In W.
Brun, G. Keren, G. Kirkebøen & H. Montgomery (Eds.), Perspectives on thinking,
judging, and decision making (pp. 21-35). Oslo: Universitetsforlaget.
Fox, C. R., Ülkümen, G., & Malle, B. F. (2011). On the dual nature of uncertainty: Cues from
natural language. Working paper. UCLA Anderson School.
Gigerenzer, G., Hertwig, R., van den Broek, E., Fasolo, B., & Katsikopoulos, K. V. (2005).
"A 30% chance of rain tomorrow": How does the public understand probabilistic
weather forecasts? Risk Analysis, 25, 623-629. doi: 10.1111/J.1539-
6924.2005.00608.X
36
Gillies, D. (2000). Philosophical theories of probability. London: Routledge.
Goodman, J. K., Cryder, C. E., & Cheema, A. (2013). Data collection in a flat world: The
strengths and weaknesses of Mechanical Turk samples. Journal of Behavioral
Decision Making, 26, 213-224. doi: 10.1002/Bdm.1753
Hacking, I. (1975). The emergence of probability: A philosophical study of early ideas about
probability, induction and statistical inference. Cambridge, UK: Cambridge
University Press.
Haddad, H., Morton, T., & Rabinovich, A. (2011). The roles of 'tone of voice' and uncertainty
when communicating climate change information. Paper presented at the British
Psychological Society Social Section Conference, Cambridge, UK.
Halkjelsvik, T., & Jørgensen, M. (2012). From origami to software development: A review of
studies on judgment-based predictions of performance time. Psychological Bulletin,
138, 238-271. doi: 10.1037/a0025996
Hogarth, R. M. (2010). Intuition: A challenge for psychological research on decision making.
Psychological Inquiry, 21, 338-353. doi: 10.1080/1047840x.2010.520260
Juanchich, M., Teigen, K. H., & Villejoubert, G. (2010). Is guilt ‘likely’or ‘not certain’?
Contrast with previous probabilities determines choice of verbal terms. Acta
Psychologica, 135, 267-277. doi: 10.1016/j.actpsy.2010.04.016
Kahneman, D., & Tversky, A. (1982). Variants of uncertainty. Cognition, 11, 143-157. doi:
10.1016/0010-0277(82)90023-3
Kaltenböck, G. (2003). On the syntactic and semantic status of anticipatory it. English
Language and Linguistics, 7, 235-255.
Keren, G., & Teigen, K. H. (2001). Why is p = .90 better than p = .70? Preference for
definitive predictions by lay consumers of probability judgments. Psychonomic
Bulletin & Review, 8, 191-202. doi: 10.3758/Bf03196156
37
Lagnado, D. A., & Sloman, S. A. (2004). Inside and outside probability judgment. In D. J.
Koehler & N. Harvey (Eds.), Blackwell handbook of judgment and decision making
(pp. 157-176). Oxford: Blackwell Publishing Ltd.
Neace, W. P., Deer, K., Michaud, S., & Bolling, L. (2011). Uncertainty is psychologically
uncomfortable: A theoretic framework for studying judgments and decision making
under uncertainty and risk. In R. Yazdipour (Ed.), Advances in entrepreneurial
finance: With applications from behavioral finance and economics (pp. 93-117). New
York: Springer.
Nordbye, G. H. H., & Teigen, K. H. (2014). Responsibility judgments of wins and losses in
the 2013 Chess Championship. Judgment and Decision Making, 9(4), 335-348.
Oppenheimer, D. M., Meyvis, T., & Davidenko, N. (2009). Instructional manipulation
checks: Detecting satisficing to increase statistical power. Journal of Experimental
Social Psychology, 45, 867-872. doi: 10.1016/J.Jesp.2009.03.009
Paolacci, G., Chandler, J., & Ipeirotis, P. G. (2010). Running experiments on Amazon
Mechanical Turk. Judgment and Decision Making, 5, 411-419.
Preacher, K. J., & Hayes, A. F. (2004). SPSS and SAS procedures for estimating indirect
effects in simple mediation models. Behavior Research Methods, Instruments, &
Computers, 36(4), 717-731. doi: 10.3758/BF03206553
Roese, N. J., & Vohs, K. D. (2012). Hindsight bias. Perspectives on Psychological Science, 7,
411-426. doi: 10.1177/1745691612454303
Taleb, N. N. (2010). The black swan: The impact of the highly improbable. New York:
Random House Trade Paperbacks.
Teigen, K. H. (1988). The language of uncertainty. Acta Psychologica, 68, 27-38. doi:
10.1016/0001-6918(88)90043-1
38
Teigen, K. H., & Brun, W. (1999). The directionality of verbal probability expressions:
Effects on decisions, predictions, and probabilistic reasoning. Organizational
Behavior and Human Decision Processes, 80, 155-190. doi: 10.1006/Obhd.1999.2857
Windschitl, P. D., & Weber, E. U. (1999). The interpretation of "likely" depends on the
context, but "70%" is 70% - Right? The influence of associative processes on
perceived certainty. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 25, 1514-1533. doi: 10.1037/0278-7393.25.6.1514
39
Tables
Table 1. Mean agreement ratings (1-7; standard deviations in parentheses) of statements in
four conditions, Experiment 1.
Internal focus(I am X% certain)
External focus (It is X% certain)
Main effect (External/Internal)
90% 70% 90% 70%F (1,221) p η2
p
Pre-outcome ratings
Knowledgeability 4.51 (0.96) 3.92 (0.92) 4.80 (1.09) 4.36 (1.10) 7.132 .008 .031
Trustworthiness 4.79 (1.03) 4.01 (0.99) 4.83 (1.13) 4.43 (1.19) 2.590 .109 .012
Informativeness 4.20 (0.97) 3.58 (1.04) 4.43 (1.20) 4.14 (1.25) 6.807 .010 .030
Post-outcome ratings
Surprise 4.09 (1.01) 3.56 (1.39) 4.16 (1.54) 3.68 (1.24) 0.332 .565 .002
Accuracy 3.36 (0.91) 3.24 (1.03) 3.67 (1.42) 3.52 (1.18) 3.726 .055 .017
Future trust 3.34 (0.99) 3.03 (1.10) 3.51 (1.42) 3.40 (1.20) 2.829 .094 .013
40
Table 2. Mean estimates of personal certainty (internal focus) vs. impersonal probability
(external focus), Experiment 2 (standard deviations in parentheses).
Scenario
Internal focus
(n = 66)
External focus
(n = 69) F(1,133) p η2p
1 – Migraine 69.05 (17.23) 60.25 (19.08) 7.889 .006 .056
2 – Exam 72.08 (21.21) 62.64 (22.04) 6.416 .012 .046
3 – Guilt 59.51 (18.30) 59.71 (14.01) .005 .945 .000
41
Table 3. Mean ratings of speakers based on their internal and external statements (standard
deviations in parentheses), Experiment 3.
Internal focus
(I am X% certain)
External focus
(There is a X% probability)
High certainty
(n = 24)
Low certainty
(n = 25)
High certainty
(n= 28)
Low certainty
(n = 23)
Supportive 4.59 (1.62) 4.57 (1.33) 3.94 (1.61) 3.88 (1.46)
Knowledgeable 4.31 (1.07) 4.25 (1.02) 4.29 (1.67) 4.00 (1.34)
Gut feelings 4.53 (1.57) 4.60 (1.28) 4.28 (1.78) 4.67 (1.65)
42
Table 4. Correlations between interest scores and agreement with certainty statements,
Experiment 5B.
Internally-focused statements
(“I am certain that…”)
(n = 44)
Externally-focused
statements
(“It is certain that…”)
(n = 43)
Interest in the Olympics r = .451, p = .002 r = -.068, p = .67
Intentions of watching the
games
r = .371, p = .013 r = -.063, p = .69
Combined interest score r = .421, p = .004 r = -.067, p = .67
Note. Only 87 of 101 participants completed the interest scales.
43